Array Camera Image Stitching Using Multi-Plane Depth Maps

The patent segments the image space into multiple depth planes based on calculated depth maps. Instead of treating the entire scene as a single plane, the method divides it into foreground and background planes at different distances from the camera. This segmentation allows independent alignment correction for each plane, resolving the internal parallax problem while maintaining a manageable processing approach.

The patent introduces depth information as an additional dimension to the traditional 2D stitching process. By calculating depth maps and creating multiple stitched images at different depth planes, the method transforms the problem from a single-plane 2D alignment task into a multi-plane 3D-aware stitching process, enabling accurate alignment of elements at different distances.

The patent performs preliminary depth map calculation and plane segmentation before the actual stitching operation. By pre-processing the images to identify depth planes and prepare alignment corrections in advance, the method enables faster execution during the stitching phase. The computationally intensive depth analysis is done once, allowing subsequent stitching operations to proceed more efficiently.

The patent applies stitching corrections selectively to different depth planes rather than uniformly across the entire image. By focusing computational resources on correcting specific planes where internal parallax occurs, rather than over-processing the entire image, the method achieves high alignment precision while avoiding unnecessary computational overhead.